Skip to main content
. 2016 Jan 26;11(1):e0147261. doi: 10.1371/journal.pone.0147261

Table 7. Performance across variations in measurement error.

Type of error Model specification Method MAD NOTFront PU20% PO20% rs
Additive measurement error θi ∼ unif(0,1), vi ∼ N(0, 0.022) DEA 0.069 11.8% 0.1% 44.2% 0.862
rDEA 0.022 2.7% 0.2% 9.8% 0.963
rSDF-CD 0.095 0.0% 32.2% 26.4% 0.797
ENS 0.048 0.0% 9.6% 20.6% 0.940
θi ∼ unif(0,1), vi ∼ N(0, 0.082) DEA 0.072 12.0% 0.8% 45.7% 0.851
rDEA 0.028 2.8% 1.8% 15.3% 0.953
rSDF-CD 0.102 0.0% 35.0% 25.5% 0.779
ENS 0.054 0.0% 13.8% 21.1% 0.929
Multiplicative measurement error θi ∼ unif(0,1), vi ∼ N(0, 0.022) DEA 0.068 11.8% 0.0% 42.6% 0.862
rDEA 0.026 2.7% 2.0% 7.4% 0.953
rSDF-CD 0.106 0.0% 50.3% 10.5% 0.762
ENS 0.056 0.0% 26.5% 6.3% 0.935
θi ∼ unif(0,1), vi ∼ N(0, 0.082) DEA 0.071 11.9% 0.4% 42.7% 0.847
rDEA 0.043 2.8% 12.4% 9.3% 0.936
rSDF-CD 0.109 0.0% 51.1% 10.9% 0.759
ENS 0.066 0.0% 34.1% 6.5% 0.926
Mixed measurement error θi ∼ unif(0,1) DEA 0.068 11.8% 0.0% 42.7% 0.859
rDEA 0.030 2.8% 3.3% 8.0% 0.950
rSDF-CD 0.107 0.0% 50.4% 10.6% 0.762
ENS 0.058 0.0% 28.1% 6.3% 0.933

Note: Numbers in bold highlight the best outcome for each performance indicator across the alternative approaches. MAD: median absolute deviation, NOTFront: percentage of misclassified DMUs, PU20%: percentage of underestimation, PO20%: percentage of overestimation, rs: Spearman’s rank correlation.